AI triage of duodenal biopsies improves workflow

Area of Science:

• Digital pathology
• Histopathology
• Artificial Intelligence in Medicine

Background:

• Histopathology laboratories face challenges in managing increasing workloads and turnaround times.
• Efficient triaging of duodenal biopsies is crucial for timely diagnosis of gastrointestinal conditions.

Purpose of the Study:

• To develop and implement an AI-based system for triaging duodenal biopsies.
• To evaluate the impact of AI triage on reporting turnaround times within an NHS histopathology laboratory.

Main Methods:

• A rule-based automation software identified and exported duodenal biopsy slides for AI triage.
• Slides with odd case numbers were processed by AI, prioritizing those with predicted abnormalities.
• A comparative analysis was performed against the routine reporting pathway for even-numbered cases.

Main Results:

• The AI triage pathway processed 329 cases (533 slides) with an AI processing time of approximately 70 seconds per slide.
• AI classifier demonstrated high sensitivity and positive predictive values for various conditions, including coeliac disease (86.7%, 100%) and normal small bowel (99.6%, 95.7%).
• Reporting turnaround times were significantly reduced for coeliac disease (6 days vs. 10 days) and non-neoplastic abnormalities (7 days vs. 10 days) in the AI triage arm (p<0.005).

Conclusions:

• An AI-based triage system for duodenal biopsies was successfully developed and implemented in an NHS histopathology laboratory.
• The AI system achieved high diagnostic accuracy and significantly improved reporting times for specific conditions.
• This study validates the feasibility and clinical utility of localized AI tools in routine diagnostic pathology workflows.